Add support for Hermes Agent, Gemini CLI, and Cursor CLI as coding agents

[Ash-Blanc]

Feature Description

Add first-class support for the following coding agents as pluggable backends:

• Hermes Agent (https://github.com/NousResearch/hermes-agent) — full skill system, MCP, local sessions
• Gemini CLI (https://github.com/google-gemini/gemini-cli) — official Google agent CLI
• Cursor CLI (https://cursor.com/cli) — Cursor's agentic coding CLI

Motivation

lat.md users already rely on Hermes, Gemini, and Cursor for day-to-day coding. Currently they have to manually invoke them or use fragile wrappers. Native support would let lat.md orchestrate tasks across these agents cleanly, with proper context passing, tool exposure, and output handling.

Proposed Solution

• Add a unified agent abstraction + config block for the three agents
• Support for:
  • Hermes: local socket/MCP integration + skill loading
  • Gemini CLI: direct invocation with model selection
  • Cursor CLI: project-aware agent sessions
• Unified interface for delegating code edits, refactors, and research tasks
• Session management and structured output support

Example config:

[agents]
hermes = { enabled = true }
gemini = { enabled = true, model = "gemini-2.0-flash" }
cursor = { enabled = true }

Alternatives Considered

Continuing to use ad-hoc subprocess calls (fragile, no unified observability).

Scope / Effort Estimate

Medium — new provider modules, config handling, and agent router updates.

Additional Context

These are the exact coding agents the user wants supported. Hermes brings powerful skill extensibility, Gemini CLI brings strong reasoning, and Cursor CLI brings editor-grade agentic coding.